Final devoicing: Production and perception studies*

Scott Myers

University of Texas at Austin

Final devoicing is a pattern of phonological distribution in which both voiced and

voiceless obstruents occur in a language, but at the end of a particular prosodic domain

(Selkirk 1978, 1986) only voiceless obstruents occur. 1, 2, 3 There are examples from all

over the world, involving both phonological word and syllable domains (cf. Passy 1891:

160; Grammont 1933: 365; Locke 1983: 118; Lombardi 1995, 1999; Blevins 2006;

Harris 2009):

* I would like to thank the following people for their helpful comments on this work: Juliette Blevins, Shigeto Kawahara, two anonymous reviewers, and audiences at the University of Massachusetts at Amherst and the Acoustical Society of America meeting in Portland. Thanks also to Lisa Selkirk, who I was lucky to have as my PhD supervisor, for her mentorship, her friendship, and the example she has always provided of elegant and insightful inquiry into how language works. 1 By "voiced" obstruents I mean obstruents that are distinguished from "voiceless" ones by having a higher proportion of vocal fold pulsing and a lower proportion of aperiodic noise. They are generally distinguished by other phonetic properties as well (Lisker 1986, Kingston and Diehl 1994), and in some of these cases periodic pulsing may not be the primary cue (Jessen and Ringen 2002). 2 Descriptions of the final devoicing pattern are often frustratingly vague. The position of neutralization is sometimes simply described as “final”, without making clear what domain it is final in (e.g. Dambriunas et al. 1966: 17), and it is often not clear whether the author checked other possible domains (e.g. whether devoicing occurs at the end of a word if the word is non-final in the phrase). Most of the descriptions cited here are also based on transcriptions, so they are inherently vague as to whether the devoicing effect is gradient or categorical. 3 The neutralization of the voicing contrast in final position in some languages is incomplete, i.e. there are measurable and perceptible differences between alternating and nonalternating final voiceless consonants (Dinnsen and Charles-Luce 1984, Port and O’Dell 1985, Slowiaczek and Dinnsen 1985, Charles-Luce and Dinnsen 1987, Slowiaczek and Szymanska 1987, Warner et al. 2004, and Dmitrieva 2005). Ernestus and Baayen (2006: 47) suggest that words with alternating final voiceless obstruents are influenced in production by the activation of the corresponding voiced-final items in their paradigm.

(1) (a) All word-final obstruents are voiceless.

• Slavic: Russian (Padgett 2002), Czech (Heim 1976: 14), Slovak (Rubach

1993: 283), Bulgarian (Scatton 1984: 20), Polish (Rubach 1984: 206)

• Romance: Walloon (Francard and Morin 1986: 457), Friulian (Baroni and

Vanelli 2000: 27), Old French (Ewert 1933: 75, 97), Ferrarese Italian

(stops only - Dinnsen and Eckman 1978: 5)

• Germanic: Dutch (Booij 1995: 22), German (Jessen and Ringen 2002),

Gothic (fricatives only -Wright 1899: 62-67; Hock 1991: 43), Old English

(fricatives only - Hock 1991: 43)

• Saranda Ekklisies Greek (stops only - Newton 1972: 103)

• Sanskrit (Whitney 1879: 46)

• Daragözu Arabic and Maltese Arabic (Abu Mansour 1996: 213, 215)

• Uyghur (Hahn 1991: 84)

• Fur (Jakobi 1990: 35)

• Luo (Tucker 1994: 35)

• Afar (Bliese 1981: 242)

• Basque (Hualde 1991: 13)

(b) All syllable-final obstruents are voiceless.

• Takelma (Sapir 1990: 35)

• Wintu (Pitkin 1984: 26)

• West Tarangan (Nivens 1992: 147)

• Romansch (Montreuil 1999: 531)

• Catalan (Hualde 1992: 393)

• Breton (Krämer 2000: 641)

• Haisla (Bach 1996: 5)

• Ron (Jungraithmayr 1970: 21)

• Malay (Ahmad 2005: 55)

• Turkish (stops only - Clements and Keyser 1983: 59-60)

• Buriat (Poppe 1960: 10)

• Efik (Cook 1969: 36)

• Manipuri (Singh 2000: 13-16)

• Thai (Iwasaki and Ingkaphirom 2005: 4)

• Vietnamese (Thompson 1965: 23)

• Various Sino-Tibetan languages (Thurgood and LaPolla 2003)

An optional pattern of word-final obstruent devoicing has been reported as a

distinguishing characteristic of a number of English dialects, e.g. those of some African

American communities (Wolfram 1969: 51, Luelsdorf 1975: 42), the Appalachian region

(Wolfram and Christian 1976: 63), or the Maori people in New Zealand (Holmes 1996).

The pattern is also reflected in language acquisition. Already in prelinguistic

babbling voiceless consonants greatly outnumber voiced consonants in utterance-final

position (Oller et al. 1976). When children begin to produce words, they often have a

stage where they produce both voiced and voiceless obstruents, but in word-final position

only voiceless ones (Velten 1943: 283, Smith 1973: 37, Smith 1979: 22, Flege 1982).

Persistence of this error type later into the acquisition process is recognized as a speech

disorder (Hodson and Paden 1981: 371, Cutts and Jensen 1983, Ingram 1989: 115).

Systematic devoicing of final voiced target consonants is also a prevalent error type in

second language acquisition, even in cases where final devoicing is not a characteristic of

either the target language or the learner’s native language (Eckman 1981, Flege and

Davidian 1984, Flege, Munro and Skelton 1992, Major and Audree 1996, Broselow,

Chen, and Wang 1998).

Scholars have long related this phonological pattern of final devoicing to the

phonetics of prepausal position (e.g. Sievers 1901: 289-290; Jespersen 1926: 101;

Bloomfield 1933: 373; Lindblom 1983: 237). There is no vocal fold vibration during

pause, so devoicing in prepausal position can be seen as assimilation to this voiceless

state (Lightner 1972: 332-333, Ingram 1989: 35). Non-speech breathing is characterized

by a wide glottal aperture to facilitate air passage, and speakers begin spreading the vocal

folds in anticipation before they are done producing speech (Sweet 1877: 65; Lisker et al.

1969: 1545; Klatt and Klatt 1990, Shadle 1997: 42; Jessen 1998; Slifka 2006). In addition

to this coarticulatory effect of the transition from speech to nonspeech, voicing in

utterance-final position is also hampered by a decline in subglottal pressure over the

course of the utterance (Westbury and Keating 1986: 156). The result is a gradual

breakdown in voicing as one approaches pause, often passing through nonmodal voicing

before a final voiceless interval. Such utterance-final devoicing has been found in

instrumental acoustic studies in English (Haggard 1978, Docherty 1992, Smith 1997),

French (Smith 1999, 2003), Finnish (Lehtonen 1970: 45, Myers and Hansen 2007), and

Kinyarwanda (Myers 2005), and noted as well in many transcription-based studies (e.g.

Michelsen’s 1988 study of the Lake Iroquoian languages). Oller and Smith (1977) found

utterance-final vowel devoicing to be a regular feature of prelinguistic babbling.

I would propose that this coarticulatory utterance-final devoicing is the initial

impetus for a sound change that results in phonological word-final devoicing. The first

step of this transformation would be that utterance-final devoicing affects the perception

of voicing contrasts in utterance-final position, inducing a tendency among listeners to

identify utterance-final obstruents as voiceless. This would be expected since utterance-

final devoicing diminishes voicing during the constriction period, a demonstrated

perceptual cue for voicing (Raphael 1971, Wolf 1978, Smith 1979, Hogan and Roszypal

1980, Kingston and Diehl 1994). As Blevins (2006) points out, lengthening of the

utterance-final consonant (Lindblom 1968) could have the same perceptual effect, since

listeners are also sensitive to the fact that voiceless obstruents are longer than

corresponding voiced ones (Denes 1955). Listeners are generally adept at compensating

in perception for coarticulatory effects (Lindblom and Studdert-Kennedy 1967, Mann and

Repp 1980), but they can fail to do so, leading to hypocorrection (Ohala 1981, 1993). In

this case, the result of failing to completely compensate for the devoicing effect of

utterance-final position would be a tendency to identify utterance-final obstruents as

voiceless.

The second step of the sound change would be that the listener generalizes over

the voicing categories he or she has identified in this way, and concludes that obstruents

in this position are voiceless. This generalization is a phonological restriction on category

distribution. Pierrehumbert (2001: 152) has shown how in an exemplar-based model even

a small bias in identification of this sort can over time lead to such a neutralization in

contrast between two speech sound categories. The extension of the pattern from

utterance-final to word- and syllable-final positions would be an analogical extension

based on the fact that every utterance-final consonant is also word- and syllable-final

(Ewert 1933: 75; Westbury and Keating 1986: 161; Hock 1991: 239).

The third step of the sound change is the spread of the pattern from the individual

innovators to a broader speech community (Labov 2001). This step depends heavily on

the innovators' relations to other speakers and the dynamics of group identity (Wedel and

Van Volkinburg 2009).

In this account, phonological final devoicing is the end result of a diachronic

process of phonologization (Hyman 1976), building on phonetic utterance-final

devoicing. It is a hypocorrective sound change (Ohala 1981, 1993), beginning with a

listener's failure to compensate perceptually for an effect of context on production.

Such a diachronic account would explain a number of the properties of

phonological final devoicing. The phonological pattern is common and has emerged

independently in numerous language groups because it results from a straightforward

change based on a pervasive pattern of laryngeal coarticulation. Voiced obstruents are

subject to change in final position because anticipation of the open glottis of nonspeech

breathing diminishes the voicing that contributes to distinguishing those sounds. Voiced

obstruents change to voiceless obstruents, because that is what a partially devoiced

obstruent tends to be mistaken for. The pattern is restricted to obstruents, because

partially devoiced sonorants are so low in intensity that they are mistaken for silence

rather than for a voiceless segment (Myers and Hansen 2007). The sound change is

recapitulated in first- and second-language acquisition because the identification errors

that are the basis for the sound change are more frequent in inexperienced learners of the

sound system.

The basic phonetic prerequisites for the phonological final devoicing pattern are

widespread, perhaps universal. But not every language that has the phonetic pattern ends

up with the corresponding phonological pattern. This is because the phonetic pattern of

utterance-final devoicing is only the first step of the change, and the emergence of

phonological final devoicing depends on all the subsequent steps as well. A listener has

to make an identification error due to the phonetic pattern often enough that it serves as

the basis of a generalization about voicing categories, and then this innovated

phonological pattern has to spread beyond that individual to a speech community. A

sound change will only occur when these events happen to line up in the right way, but

the point is that this series of events is more likely than one with a less frequent starting

point (Yu 2004).

The articulatory and acoustic bases of this diachronic account are well-supported;

the studies cited above demonstrate the effect of utterance-final position on the actions of

the vocal folds and on the resulting acoustic reflexes of voicing. But no evidence has

been provided to date for the claim that these acoustic effects of utterance-final

coarticulation affect identification of voicing categories and lead to a tendency to identify

utterance-final obstruents as voiceless. The aim of the present study is to test this claim.

English was chosen as the language of the study, since it has a robust contrast in

voicing in word-final position (e.g. pat/pad), and English speakers are therefore

experienced in producing and perceiving contrasting voice categories in this position.

The first experiment is a production study, which is meant both to explore the conditions

for the utterance-final devoicing effect and to generate stimuli for the perception

experiments. The second and third experiments are perception experiments, in which

English-speaking listeners identify words belonging to minimal pairs differing in final

voicing (e.g. proof/prove), excised from utterance-final or nonfinal position.

1. Experiment #1: Production

1.1 Methods

The test items, listed in (2), all belonged to minimal pairs differing just in the

voicing of a word-final obstruent: final voiceless vs. final voiced fricative (e.g.

proof/prove), or final voiced vs. voiceless stops (e.g. greet/greed). There were ten test

words for each class of word-final segment, for a total of 40 items.

(2) Test words

(a) Fricative-final (voiceless - voiced) (b) Stop-final (voiceless - voiced)

loose - lose greet - greed

proof - prove feet - feed

cease - seize beat - bead

leaf - leave seat - seed

Bruce - bruise loop - lube

noose - news neat - need

belief - believe moot - mood

grief - grieve leak - league

relief - relieve heat - heed

use (noun) - use (verb) sweet - Swede

To control for the effects of stress (Lehiste 1970: 36) and vowel height (Lindblom 1968)

on the duration of the vowel, the final syllable in all test words bore main word stress and

had a high tense vowel.

Each test word occurred in two carrier sentences. To control for the effect of

utterance length on vowel duration (Lindblom 1968), each carrier sentence consisted of

15 syllables. In one sentence, the test word was sentence-final (e.g. The garage can

tighten any of the bolts that are too loose), while in the other it was non-final and

preceding a word beginning with a nasal stop (e.g. There is a loose nylon cover over the

whole area). Each of the 40 test words occurred in 2 sentence positions, so there were 80

test sentences for each speaker.

6 adult native speakers of American English produced the materials.4 The test

sentences were presented in random order at 5-second intervals in a timed Powerpoint

presentation on a laptop, interspersed with 60 distractor sentences (stimuli for another

experiment). The speaker read them aloud seated in a sound recording booth and was

recorded on a solid-state digital recorder. 480 test sentences were produced by the 6

4 The subjects were from Oklahoma, Illinois, Tennessee, Oregon, and California.

speakers, but of those 1 item was excluded because the speaker produced the wrong

word, and 2 were excluded due to background noise that made the measurements

impossible. That left 477 items for analysis.

There are many acoustic correlates for voicing (Lisker 1986), but since this study

concerns the interaction of such correlates with utterance position, we focus on measures

which are defined for both stops and fricatives, and which are known to be sensitive to

both voicing and utterance position.

The duration of the following intervals was measured, using Praat

(http://www.fon.hum.uva.nl/praat/): (a) the vowel, (b) the constriction for the postvocalic

consonant (if any), (c) the release (if any), and (d) the voiced and voiceless subintervals

within the VC sequence (a) -(c).

The onset of the vowel was defined as the onset of an increase in amplitude and

wave complexity after the prevocalic consonant. The offset of the vowel was defined as

the end of F2 and F3 of the vowel. The constriction interval was a period of minimum

amplitude beginning with the vowel offset. In the case of stops this corresponded to the

closure interval, while in fricatives it was the noise interval. The release was the interval

of increased amplitude after the constriction and before the onset of a following sound, if

any. In a released stop, this was the noise burst and aspiration. In fricatives, this was a

period characterized by a shift to lower frequency and lower intensity noise. In these

recordings, all utterance-final stops were released, suggesting a fairly careful

pronunciation.

The voiced interval was from the onset of the vowel to the offset of quasiperiodic

pulses in the waveform. The voiceless interval consisted either of noise without such

pulsing, or silence (in a stop closure). In these recordings, the voiced interval was always

continuous, i.e. all the VC sequences measured had an initial voiced interval, which was

followed in some cases by a voiceless interval that stretched to the end of the VC

sequence. The transition from voiced to voiceless always occurred either in the

constriction interval or in the release.

The constriction interval was expected to be longer in voiceless obstruents than in

voiced obstruents (Denes 1955, Lisker 1957, Stevens et al. 1992), and longer in

utterance-final than in nonfinal position (Byrd et al. 2005).

The duration of the voiceless interval within the consonant (constriction + release)

was expected to be longer in a voiceless than in a voiced consonant (Raphael 1971,

Kingston and Diehl 1994), and longer in utterance-final than in nonfinal position

(Haggard 1978, Docherty 1992, Smith 1997). The absolute duration of the voiceless

interval was chosen in this study as a measure rather than the proportion of the voiced

interval to the whole consonant interval because the latter measure had a bimodal

distribution, complicating statistical analysis (cf. Kuzla, Cho and Ernestus 2007). The

duration of the voiceless interval was chosen over that of the voiced interval since it is a

devoicing effect that we are aiming to measure.

The duration of the vowel was expected to be longer before voiced than before

voiceless consonants (Chen 1970), and longer in utterance-final position than in nonfinal

position (Lindblom 1968, Oller 1973). The ratio of the vowel duration to the duration of

the whole vowel + consonant sequence has been argued to be a more robust acoustic

correlate of voicing than the absolute duration of the vowel alone (Kohler 1979, Barry

1979, Pind 1986). This V/VC ratio is greater when C is voiced than when it is voiceless,

because the vowel before a voiced coda is longer and the coda itself is shorter. The ratio

is also lower in utterance-final syllables than in nonfinal syllables (Barry 1979), because

final lengthening is gradient and has more of an effect on the utterance-final coda than on

the preceding vowel, proportionally (Turk 1999).

In the statistical analysis a mixed model was used in which speaker and test word

were treated as random effects, and fixed effects were voicing (voiced/voiceless), manner

(stop/fricative), and (utterance) position (final/nonfinal). The alpha level was .05. A given

acoustic measure was considered to show utterance-final devoicing when a significant

effect of utterance-final position coincided with (i.e. went in the same direction as) the

effect of having a voiceless obstruent in coda position.

1.2 Results

1.2.1 Constriction duration

Fig. 1 presents the percentile distribution of constriction duration by manner,

voicing and position:

Fig. 1: Constriction duration (ms) by manner, voicing and position

Mean constriction duration was greater in voiceless obstruents (118 ms) than in voiced

obstruents (77 ms), greater in fricatives (138 ms) than in stops (57 ms), and greater in

utterance-final syllables (124 ms) than in nonfinal ones (71 ms). All three main effects

were significant (d.f. = 1, 469): voicing (F = 109.0, p < .001), manner (F = 444.5, p <

.001), and position (F = 433.4, p < .001). The 2-way interactions were also significant

(d.f. = 1, 469): voicing*position (F = 30.7, p < .001), voicing*manner (F = 36.1, p <

.001), and manner*position (F = 274.7, p < .001). The 3-way interaction was not

significant.

In order to explore the interactions, the data was split into subsets according to

manner and position. Considering the manner classes, both the voicing and position main

effects were significant in both stops and fricatives: voicing in stops (F (1, 234) = 17.0, p

< .001), voicing in fricatives (F (1, 235) = 96.8, p < .001), position in stops (F (1, 234) =

13.1, p < .001), and position in fricatives (F (1, 235) = 612.1, p < .001). The interaction

between the two factors was significant in both stops (F (1, 234) = 17.5, p < .001) and

fricatives (F (1, 235) = 17.4, p < .001).

Splitting the data according to position class, voicing was a significant main effect

in both final and nonfinal positions (final, F (1, 233) = 190.9, p < .001; nonfinal, F (1,

236) = 35.2, p < .001). The main effect of manner was also significant in both position

classes (final, F (1, 233) = 983.4, p < .001; nonfinal (1, 236) = 78.7, p < .001), as was the

interaction of voicing and manner (final, F (1, 233) = 41.6, p < .001; nonfinal (1, 236) =

23.1, p < .001).

In both stops and fricatives, and in both final and nonfinal position, constriction

duration was significantly longer in voiceless than in voiced obstruents, as previously

found by Denes (1955). As in Byrd et al. (2005), the constriction duration was longer in

utterance-final than in non-final position. The two factors interacted so that the difference

between the voiced and voiceless group means was greater in final position (55 ms) than

in nonfinal position (27 ms). The lengthening of constriction duration associated with

utterance-final position coincided with and reinforced the lengthening effect of belonging

to the voiceless category, so with respect to this measure utterance-final position had a

devoicing effect. The same result is obtained if the whole consonant duration

(constriction + release) is used as a measure, instead of just the constriction interval.

1.2.2 Duration of the voiceless interval

Fig. 2 presents the percentile distribution of the duration of the voiceless interval

by manner, voicing, and position class.

Fig. 3. Duration of the voiceless interval by manner, voicing and position

The mean duration of the voiceless interval was greater in voiceless obstruents (146 ms)

than in voiced ones (78 ms), greater in fricatives (130 ms) than in stops (93 ms), and

greater in final position (181 ms) than in nonfinal position (43 ms). All three main effects

are significant (d.f. = 1, 469): voicing (F = 194.8, p < .001), manner (F = 56.9, p < .001)

and position (F = 1213.4, p < .001). All of the interactions except that of voicing and

position were significant (d.f. = 1, 469): voicing and manner (F = 5.0, p = .03), manner

and position (F = 7.8, p = .01), and voicing*manner*position (F = 4.4, p = .04).

In the manner subsets, the main effect of voicing was significant (stops, F (1, 234)

= 45.5, p < .001; fricatives, F (1, 235) = 272.4, p < .001), as was the main effect of

position (stops, F (1, 234) = 406.8, p < .001; fricatives, F (1, 235) = 970.3, p < .001). The

interaction of voicing and position was significant in stops (F (1, 234) = 5.0, p = .03), but

not fricatives.

In the position subsets, the main effect of voicing was significant (final, F (1, 233)

= 108.0, p < .001; nonfinal, F (1, 236) = 104.5, p < .001), and so was the main effect of

manner (final, F (1, 233) = 44.8, p < .001; nonfinal, F (1, 236) = 17.4, p < .001). The

interaction of voicing and manner was significant in the nonfinal subset (F (1, 236) = 9.7,

p = .002), but not in the final subset.

Thus the voiceless interval in the consonant was significantly longer in voiceless

obstruents than in voiced ones in both stops and fricatives, and both final and nonfinal

position. The voiceless interval was longer in utterance-final position than in nonfinal

position. The voicing and position factors interacted in such a way that the difference

between the voiced and voiceless group means was greater in final position (74 ms) than

in nonfinal position (63 ms). For this measure, the lengthening effect of final position

coincided with the lengthening effect of a voiceless obstruent, so utterance-final position

had a devoicing effect. The same general pattern holds if the proportion of voicing to

consonant duration is used as a measure.

1.2.3 Vowel duration

Fig. 3 presents the percentile distribution of vowel duration (ms) by coda voicing

and utterance position:

Fig. 3: Vowel duration (ms) by manner, voicing and utterance position

Mean vowel duration was greater before a voiced obstruent (199 ms) than before

voiceless ones (mean = 152 ms), slightly greater before a fricative (177 ms) than before a

stop (175 ms), and greater in an utterance-final syllable (198 ms) than in a nonfinal

syllable (154 ms). The main effects of voicing and position were significant (d.f. = 1,

469): voicing, F = 57.8, p < .001; position, F = 255.0, p < .001, but not the main effect of

manner (F < 1). The significant interactions (d.f. = 1, 469) were those of voicing and

position (F = 21.0, p < .001), voicing and manner (F =13.3, p < .001), and voicing,

manner and position (F = 18.8, p < .001). The interaction of manner with voicing was not

significant (F < 1).

The main effect of voicing was significant in both manner subsets (stops, F (1,

234) = 22.7, p < .001; fricatives, F (1, 235) = 35.6, p < .001), as was the main effect of

position (stops, F (1, 234) = 65.5, p < .001; fricatives, F (1, 235) = 226.3, p < .001). The

interaction between voicing and position was significant in fricatives (F (1, 235) = 46.7, p

< .001), but not in stops (F < 1).

Examining the final and nonfinal subsets separately, the main effect of voicing

was significant in both the final and nonfinal subsets (final, F (1, 233) = 79.7, p < .001;

nonfinal, F (1, 236) = 19.2, p < .001), while the main effect of manner was not significant

in either subset. The interaction of voicing and manner was significant in the final subset

(manner, F (1, 233) = 6.9, p = .009), but not in the nonfinal subset.

Thus, as in previous studies (e.g. Chen 1970), vowel duration in this sample was

significantly greater before a voiced consonant than before a voiceless one in stops and

fricatives, and in final and nonfinal position. As in previous studies (e.g. Oller 1973),

vowel duration was significantly greater in an utterance-final syllable than in a nonfinal

one. The two factors interacted in that the difference between the voiced and voiceless

group means was greater in final position (60 ms) than in nonfinal position (33 ms).

Umeda (1975) found a similar interaction in a study of connected speech in English, but

in her study the significant effects of voicing were limited to final position. For our

purposes, what matters most is that with regard to this voicing cue, the lengthening

effects of utterance-final position coincide with and reinforce the effect of a voiced coda,

rather than the voiceless category as with the previous two measures. Thus with respect

to this measure, utterance-final position has a voicing effect, not the expected devoicing

effect.

However, it should be noted that some scholars have argued that the appropriate

cue for voicing in a postvocalic consonant is not the absolute duration of the preceding

vowel, but the ratio of the vowel duration to the duration of the whole VC sequence

(Kohler 1979, Barry 1979, Pind 1986). The percentile distribution of this measure,

V/VC, is presented in Fig. 4.

Fig. 4: The ratio of vowel duration to VC duration, by manner, voicing and position class

The mean V/VC ratio was greater for voiced codas (.64) than for voiceless ones (.51),

greater with stops (.61) than with fricatives (.54), and greater in nonfinal position (.66)

than in utterance-final position (.49). All the main effects were significant for this

measure (d.f. = 1, 469): voicing (F = 129.2, p < .001), manner (F = 36.9, p < .001), and

position (F = 568.8, p < .001). The significant interactions (d.f. = 1, 469) were those of

voicing and manner (F = 14.6, p < .001), and manner and position (F = 25.8, p < .001).

The main effect of voicing was significant in both the stop subset (F (1, 234) =

19.6, p < .001) and the fricative subset (F (1, 235) = 220.3, p < .001). The main effect of

position was also significant in both manner subsets (stops, F (1, 234) = 325.9, p < .001;

fricatives, F (1, 235) = 269.0, p < .001). In neither manner subset was there a significant

interaction between voicing and position.

Dividing the dataset by position, the main effect of voicing was significant (final,

F (1, 233) = 133.6, p < .001; nonfinal, F (1, 236) = 60.4, p < .001), as was that of manner

(final, F (1, 233) = 8.5, p =.004; nonfinal, F (1, 236) = 38.6, p < .001) and the interaction

of voicing and manner (final, F (1, 233) = 14.2, p < .001; nonfinal, F (1, 236) = 7.1, p =

.008).

Thus the V/VC ratio was significantly greater with voiced codas than with

voiceless ones in both stops and fricatives, and both final and nonfinal position. The ratio

was significantly greater in final than in nonfinal position for both stops and fricatives. In

this case there was no significant interaction between these two factors, so that there was

no significant difference in the voicing effect depending on utterance position. The

lowered ratio associated with final position coincided with the lowering effect of a

voiceless coda - an utterance-final devoicing effect.

1.2.4 Discussion: Production study

It has been found that there are three important acoustic cues for voicing that

show utterance-final devoicing in this dataset, such that the effect of final position

coincides with the effect of voicelessness. Constriction duration and the duration of the

voiceless interval are longer in final than nonfinal position, and longer in voiceless than

in voiced consonants. The V/VC duration ratio is lower in final than in nonfinal position,

and lower with voiceless than with voiced codas.

Vowel duration goes the other direction, since vowels were longer in final

position and with voiced codas. There are also other cues for voicing that have not been

measured in this study (Lisker 1986), such as f0 or F1, in which the effects of final

position might coincide with those of the voiced category.

The question is which class of acoustic correlates is dominant in the identification

of voicing categories in utterance-final position. If, as we have hypothesized, listeners

tend to identify utterance-final obstruents as voiceless, this would suggest that the cues

that are most salient to listeners are those in which the effect of utterance-final position

coincides with the effect of a voiceless coda. If, on the other hand, absolute vowel

duration is the most salient acoustic correlate of coda voicing for listeners, then we would

expect that listeners would tend to identify utterance-final obstruents as voiced. The

perception experiments presented in the next session will provide evidence on this point.

2. Experiments #2a and #2b: Perception

We have seen in the production study that utterance-final position has significant

effects on the acoustic correlates of the voicing contrast in English. The goal of the

perception studies is to investigate what consequences, if any, these acoustic effects have

on listeners' identification of categories contrasting in voicing.

2.1 Methods

The 20 obstruent-final voiced-voiceless pairs in (2) above served as the test items

for the perception experiments. The test words were excised from the soundfiles

produced in the first experiment, cutting at the closest zero crossing preceding the onset

of the word and the closest zero crossing immediately following the end of the word

(including the final consonant release, if any).

There were 477 tokens from the production study. 2 of these were excluded due to

speaker error, leaving 475 stimuli. The recordings were processed in Adobe Audition.

They were normalized to the same peak intensity (-15 dB relative to full scale). To

decrease the abruptness of the soundfile onset, a 100 ms. interval of silence was added to

the beginning of each soundfile, and if the initial sound wasn’t a plosive, the initial

portion was reset to fade in gradually.

In the stimuli for Experiment 2a, the files were presented in the clear, without any

added noise. However, it was expected that the error rate would be low in this case, and

perhaps too low to provide enough information about the kinds of errors listeners were

prone to. Thus a new series of stimuli was created by taking the stimuli for Experiment

2a and mixing in pink noise (in which intensity is inversely proportional to frequency) at

a signal-to-noise ratio of 10 dB. These stimuli were used in Experiment 2b.

Since no changes were made to f0, voice quality, or segment duration, the stimuli

do not sound like isolation words, i.e. complete intonational phrases consisting of a single

word, but are clearly incomplete snippets from a longer utterance. Subjects were told that

the stimuli were cut out of longer sentences, and that that is why they might sound odd.

The stimuli were presented through headphones from a laptop computer using

Superlab (Cedrus). The stimuli were blocked by word pair, and presented in a different

random order within the block for each subject. For each block, the subject was presented

with the choice of items on the laptop screen – one choice in blue on the left side of the

screen, and the other in red on the right. Sound files were presented every 2 seconds, and

subjects were instructed to listen to each one and press a key on a response box to

indicate as quickly as possible their choice as to which word they heard. A blue key on

the left of the box corresponded to the left-hand choice in blue, and a red key on the right

corresponded to the right-hand choice in red. Voiced and voiceless choices were evenly

distributed between left and right.

16 adult native speakers of American English participated in each of the two

experiments. Each subject participated in only one of the experiments, and none of the

subjects for the production experiment participated as subjects in these perception

experiments.

Since the experiment involves native speakers of English identifying familiar

words of English, it was expected that the error rate would be quite low. But the

hypothesis was that there would nevertheless be a significant tendency among listeners to

identify utterance-final obstruents as voiceless, and that there would be more voiceless

judgements for forms excised from utterance-final position than for forms from

utterance-medial position.

The dependent variable here is a categorical response (voiced/voiceless), so the

models are based on a binomial distribution. Subject (i.e. listener), talker, and minimum

pair were included as random effects, and the fixed effects were stimulus voicing (voiced

coded as 0, and voiceless as 1) and stimulus position (final coded as 0, and nonfinal as 1).

The logistic regression models expressed the likelihood of a voiceless response based on

these factors.

2.2 Results

There were 7600 trials in each experiment (475 stimuli * 16 subjects). In

experiment 2a there were 67 nonresponses, i.e. cases in which the subject did not press

either key in the 2-second interval allowed. These were excluded from the analysis,

leaving a total of 7533 responses. In Experiment 2b, 62 nonresponses were excluded, as

well as 2 responses with suspiciously low response times below 110 ms (from stimulus

onset), leaving a total of 7536 responses.

As expected, subjects were in general quite accurate in their identification of the

stimuli. 6844 of the responses in Experiment 2a (90.9%) were correct, while in

Experiment 2b, with added noise, 6539 responses (86.8%) were correct.

The responses are broken down by stimulus category in Table 1 for Experiment

2a and in Table 2 for Experiment 2b. Correct responses are highlighted in boldface.

Table 1: Experiment 2a: Number (and percentage) of voicing responses by stimulus

manner, position and voicing

Stimulus

manner

Stimulus

position

Stimulus

voicing

Number (and

percentage) of

voiced responses

Number (and

percentage) of voiceless

responses

Stop Nonfinal Voiced 866 (92%) 74 (8%)

Stop Nonfinal Voiceless 98 (10%) 841 (90%)

Stop Final Voiced 891 (95%) 49 (5%)

Stop Final Voiceless 72 (8%) 866 (92%)

Fricative Nonfinal Voiced 919 (97%) 33 (3%)

Fricative Nonfinal Voiceless 189 (20%) 760 (80%)

Fricative Final Voiced 820 (89%) 106 (11%)

Fricative Final Voiceless 68 (7%) 881 (93%)

Table 2: Experiment 2b: Number (and percentage) of voicing responses by stimulus

manner, position and voicing

Stimulus

manner

Stimulus

position

Stimulus

voicing

Number (and

percentage) of

voiced responses

Number (and

percentage) of voiceless

responses

Stop Nonfinal Voiced 802 (86%) 136 (14%)

Stop Nonfinal Voiceless 114 (12%) 825 (88%)

Stop Final Voiced 833 (88%) 110 (12%)

Stop Final Voiceless 83 (9%) 855 (91%)

Fricative Nonfinal Voiced 859 (90%) 93 (10%)

Fricative Nonfinal Voiceless 184 (19%) 766 (81%)

Fricative Final Voiced 767 (83%) 162 (17%)

Fricative Final Voiceless 115 (12%) 832 (88%)

In both experiments, the fricatives show a higher percentage of voiceless

responses for corresponding stimuli from final position than for those from nonfinal

position. Thus in Experiment 2a, voiceless fricative stimuli were correctly identified as

voiceless in 93% of the final cases but only 80% of the nonfinal ones. Voiced fricative

stimuli in the same experiment were incorrectly identified as voiceless in 11% of the final

cases, as compared to 6% of the nonfinal ones. The same generalization held for the

fricative stimuli in Experiment 2b.

However, the stops did not show such a pattern. In both experiments, as with

fricative stimuli, voiceless stops were correctly identified as voiceless more often in final

position than in nonfinal position. For example, among stimuli with voiceless stops in

Experiment 2a, 92% were identified as voiceless in final position, compared to 90% in

nonfinal position. But, unlike the fricative stimuli, the incorrect identification of voiced

stops as voiceless was slightly less frequent in final position than in nonfinal position. For

example, in Experiment 2a, 8% of the final voiced stops were identified as voiceless,

compared with 10% of the nonfinal voiced stops.

The results of the statistical analysis are given in Table 3 for Experiment 2a, and

in Table 4 for Experiment 2b.

Table 3. Experiment 2a: Fixed effects

Fixed effect Estimated

coefficient

z p

Voicing 5.2 29.0 <.001

Position -1.4 -6.4 <.001

Manner -0.9 -3.3 <.001

Voicing * Position 0.1 0.4 .66

Voicing * Manner 0.9 6.4 <.001

Position* Manner 1.8 6.4 <.001

Voicing*Manner*Position -1.0 -2.6 .01

Table 4. Experiment 2b: Fixed effects

Fixed effect Estimated

coefficient

z p

Voicing 3.9 27.7 <.001

Position -0.7 -4.9 <.001

Manner -0.5 -2.2 .03

Voicing * Position 0.1 0.6 .53

Voicing * Manner 0.9 4.2 <.001

Position* Manner 1.0 4.9 <.001

Voicing*Manner*Position -0.8 -2.7 .007

In both experiments, all the main effects were significant. The positive coefficient for

voicing, in conjunction with the significant z value, indicates that a voiceless stimulus

(coded 1) had a significantly greater likelihood to be identified as voiceless than a voiced

stimulus (coded 0) had. This just reflects the high accuracy of identification of voicing

categories. The negative coefficient for position, in conjunction with the significant effect

for that factor, means that a nonfinal stimulus (coded 1) had a significantly smaller

likelihood to be identified as voiceless than a final stimulus (coded 0). The negative

coefficient for manner meant that a stop (coded 1) was less likely than a fricative (coded

0) to be identified as voiceless. However, there were significant interactions among these

factors, complicating the interpretation.

The dataset was split into stop and fricative subsets so that the effects of voicing

and position could be examined in these different manner subsets. The results of these

tests are presented in tables 5-8.

Table 5. Experiment 2a: Fricatives

Fixed effect Estimated coefficient z p

Voicing 5.3 28.1 < .001

Position -1.4 -6.5 < .001

Voicing * Position 0.2 0.6 .53

Table 6. Experiment 2b: Fricatives

Fixed effect Estimated coefficient z p

Voicing 4.0 27.1 < .001

Position -0.7 -4.9 < .001

Voicing * Position 0.1 0.6 .53

Table 7. Experiment 2a: Stops

Fixed effect Estimated coefficient z p

Voicing 6.1 27.8 < .001

Position 0.5 2.4 .02

Voicing * Position -0.8 -3.2 .001

Table 8. Experiment 2b: Stops

Fixed effect Estimated coefficient z p

Voicing 4.8 28.6 < .001

Position 0.3 2.0 .048

Voicing * Position -0.7 -3.1 .002

In the fricative subset in both experiments (Tables 5 and 6), the main effect of

voicing was significant, with a positive coefficient reflecting the subjects' largely

accurate identification of voicing categories. The main effect of position was also

significant, with a negative coefficient reflecting an association of final position with

voiceless responses and nonfinal position with voiced responses. The interaction was not

significant, so there was no evidence that the effect of final position on identification was

different for stimuli with a voiced fricative compared to those with a voiceless fricative.

In the stop subset (Tables 7 and 8), both main effects and their interaction were

significant. The coefficient for voicing was positive in both experiments, indicating that

among stops, as with fricatives, voiceless stimuli were more likely to get voiceless

responses than were voiced stimuli. But the coefficient for position is positive, and that

for the interaction is negative. This reflects the fact, noted above, that the effect of

position was different for voiced stimuli than for voiceless ones. Voiceless stops were

more often correctly identified as voiceless in final than in nonfinal position, as expected

and as found with fricatives, but voiced stops were unexpectedly more often identified as

voiceless in nonfinal than in final position.

The nonfinal condition was expected to be a neutral control condition, and in

Experiment 2b this is how it turned out, with voiced and voiceless responses evenly split.

But in Experiment 2a there were more voiced responses than voiceless responses in

nonfinal position (51% among the stops, and 58% among the fricatives). This tendency

toward voiced responses could be due to coarticulatory voicing, since each test obstruent

is between two voiced sonorants (a vowel and a nasal). But it raises the possibility that

the observed effects of position are due more to nonfinal voicing than to final devoicing.

To exclude this possibility, we restrict our view to utterance-final fricatives. In this set, in

both experiments, the proportion of voiceless responses was significantly greater than

that expected by chance: Experiment 2a, z = 1.7, p = .04; Experiment 2b, z = 2.2, p = .02.

2.3 Discussion: Perception Studies

The hypothesis was that the acoustic effects of utterance-final devoicing observed

in the production study would lead to a tendency for word-final obstruents from

utterance-final position to be identified as voiceless. The two perception experiments

have provided support for this hypothesis in fricative-final words, but not in stop-final

words, and in particular not in words ending in voiced stops.

One explanation of this difference between stops and fricatives could lie in the

effects of manner on the acoustic correlates of voicing found in the production study.

Fricatives had a longer constriction duration, a longer voiceless interval, and a lower

V/VC duration ratio than stops, and in all of these measures the effect of a fricative thus

coincided with the effect of a voiceless sound.5 As Ohala (1983: 201) pointed out,

"voiced fricatives have more exacting aerodynamic requirements than do voiced stops",

since voicing requires supralaryngeal pressure to be lower than sublaryngeal pressure,

while at the same time supralaryngeal pressure behind the oral constriction must be high

enough to generate turbulent oral airflow. Ohala suggests that this aerodynamic issue is

part of the reason that consonant inventories with only voiceless fricatives are twice as

common in Ruhlen's (1987) survey than inventories in which all stops are voiceless. He

goes on to propose that it is also the reason that "in American English the 'voiced'

fricatives /v, z/ are more likely to be devoiced in word-final position than are the stops /b,

d, g/." Thus it could be that the inherently weaker voicing cues for fricatives are more

susceptibility to confusing interference from utterance-final devoicing than the stronger

voicing cues for stops.

5 Manner had no significant effect on the last measure considered: absolute vowel duration.

3. Conclusion

The production study reported here has replicated previous findings of a

significant utterance-final devoicing effect in English. Utterance-final obstruents

displayed a longer constriction duration, a longer voiceless interval, and a lower ratio of

voswel duration to total VC duration in comparison to nonfinal obstruents. In all these

measures, the effect of utterance-final position coincides with the effect of belonging to

the voiceless category.

The two perception studies tested the hypothesis that the acoustic effects of

utterance-final position lead to a tendency to identify utterance-final obstruents as

voiceless. The results of the studies clearly support that hypothesis for fricatives, but not

for stops. Voiceless stops in final position were more likely to be correctly identified as

voiceless than those in nonfinal position, but there was no tendency in either perception

study for voiced stops to be misidentified as voiceless. I have suggested that this

difference might have been due to that the effects of fricatives on the voicing cues

coincides with the effects of voicelessness and utterance-final position.

The results provide a basis for rejecting the null hypothesis that utterance-final

position has no effect on identification of voicing categories. The fact that fricatives from

utterance-final position tended to be identified as voiceless further suggests that for word-

final fricatives the dominant perceptual cues for voicing are those like constriction

duration, voiceless interval duration, and V/VC ratio, for which the effects of final

position coincide with those of the voiceless category.

The test words in the perception experiments were presented in isolation, without

following context. Thus listeners were not able to use their abilities to compensate for

the acoustic effects of coarticulation (Lindblom and Studdert-Kennedy 1967, Mann and

Repp 1980), and to fill in missing information top-down from discourse context (Warren

1970) or from knowledge of the lexicon (Ganong 1980). These abilities are impressive,

but they are not infallible, as evidenced by the fact that listeners do make identification

errors in conversation even with the full phonetic and discourse context known. The

perception experiments reported here provide information about a baseline pattern of

errors when such contextual information is held constant by complete removal of context.

The results thus provide support, at least in the case of fricatives, for one step in

the diachronic account sketched in the introduction of how utterance-final phonetic

devoicing provides the basis for a sound change resulting in phonological final devoicing.

In this account the phonetic basis of final devoicing is limited to utterance-final position,

but the pattern is generalized from there to word-final and syllable-final position through

analogical extension. The results of the perception experiments also suggest that the

perceptual basis of the sound change might be limited to fricatives, and extended from

that subset of obstruents to the class of all obstruents.

One might expect from this that utterance-final fricative devoicing should be the

most common version of the phonological pattern of final devoicing, since it requires the

fewest further steps of generalization. There are cases of devoicing limited to fricatives

(e.g. Gothic: Wright 1899: 62-67; Hock 1991: 43) and there are cases in which the

devoicing is said to be limited to utterance-final position (e.g. some Yiddish dialects:

Wetzels and Mascaró 2001: 224; Polish: Jassem and Richter 1989; examples in Blevins

2006: 142). But it certainly does not seem as if such cases are more common than word-

final devoicing, or devoicing of all obstruents including stops.

It would appear, then, that the tendency to generalize the phonological pattern,

from utterance-final words to all words and from fricatives to all obstruents is strong

enough to render utterance-final phonological fricative devoicing unstable. If so, this

tendency must lie not in the phonetic basis of the pattern, but in how language learners

make generalizations about the distribution of speech sound categories (Hayes 1999,

Moreton 2008).

References

Abu-Mansour, M. (1996). Voice as a privative feature: Assimilation in Arabic. In M. Eid

(ed.) Perspectives on Arabic Linguistics VIII. John Benjamins, Amsterdam. 201-231.

Ahmad, Z. (2005). The Phonology-Morphology Interface in Malay: An Optimality

Theoretic Account. Pacific Linguistics, Canberra.

Bach, E. (1996). Building words in Haisla. Unpublished manuscript, University of

Massachusetts, Amherst.

Baroni, M. and L. Vanelli (2000). The relationship between vowel length and

consonantal voicing in Friulian. In L. Repetti (ed.) Phonological Theory and the

Dialects of Italy. John Benjamins, Amsterdam. 13-44.

Barry, W. (1979). Complex encoding in word-final voiced and voiceless stops. Phonetica

36. 361-372.

Blevins, J. (2006). A theoretical synopsis of Evolutionary Phonology. Theoretical

Linguistics 32. 117-166.

Bliese, L. (1981). A Generative Grammar of Afar. Summer Institute of Linguistics,

Arlington (Texas).

Bloomfield, L. (1933). Language. Holt, Rinehart, and Winston, New York.

Booij, G. (1995). The Phonology of Dutch. Oxford University Press, Oxford.

Broselow, E., S.-I. Chen, and C. Wang (1998). The emergence of the unmarked in second

language phonology. Studies in Second Language Acquisition 20. 261-280.

Byrd, D., S. Lee, D. Riggs, and J. Adams (2005). Interacting effects of syllable and

phrase position on consonant articulation. Journal of the Acoustical Society of

America 118. 3860-3873.

Charles-Luce, J. and D. Dinnsen (1987). A reanalysis of Catalan devoicing. Journal of

Phonetics 15. 187-190.

Chen, M. (1970). Vowel length variation as a function of the voicing of the consonant

environment. Phonetica 22, 129-159.

Clements, G. N. and S. J. Keyser (1983). CV Phonology. MIT Press, Cambridge.

Cook, T. L. (1969). The Pronunciation of Efik for Speakers of English. Intensive

Language Training Center, Indiana University, Bloomington.

Cutts, H. and P. Jensen (1983). Speech timing of phonologically disordered children:

Voicing contrast of initial and final stop consonants. Journal of Speech and Hearing

Science 26. 501-510.

Dambriunas, L., A. Klimas, and W. Smalstieg (1966). Introduction to Modern

Lithuanian. Franciscan Fathers, Brooklyn.

Denes, P. (1955). Effect of duration on the perception of voicing. Journal of the

Acoustical Society of America 27. 761-764.

Dinnsen, D. and J. Charles-Luce (1984). Phonological neutralization, phonetic

implementation and individual differences. Journal of Phonetics 12. 49-60.

Dinnsen, D. and F. Eckman (1978). Some substantive universals in atomic phonology.

Lingua 45. 1-14.

Dmitrieva, O. (2005). Incomplete Neutralization in Russian Final Devoicing: Acoustic

Evidence from Native Speakers and Second Language Learners. Master’s thesis,

University of Kansas, Lawrence.

Docherty, G. (1992). The Timing of Voicing in British English Obstruents. Foris, Berlin.

Eckman, F. (1981). On the naturalness of interlanguage phonological rules. Language

Learning 31. 195-216.

Ernestus, M. and R. H. Baayen (2006). The functionality of incomplete neutralization in

Dutch: The case of past-tense formation. In L. Goldstein, D. Whalen, and C.T. Best

(eds.) Laboratory Phonology 8. Berlin, Mouton de Gruyter. 27-49.

Ewert, A. (1933). The French Language. Faber and Faber, London.

Flege, J. E. (1982). English speakers learn to suppress final devoicing. In K. Tuite, R.

Schneider, and R. Chametsky (eds.) Papers from the 18th Regional Meeting. Chicago

Linguistics Society, Chicago. 111-122.

Flege, J. and R. Davidian (1984). Transfer and developmental processes in adult foreign

language speech production. Applied Psycholinguistics 5. 323-347.

Flege, J., M. Munro, and L. Skelton (1992). Production of the word-final English /t/ - /d/

contrast by native speakers of English, Mandarin, and Spanish. Journal of the

Acoustical Society of America 92. 128-143.

Francard, M. and Y.-C. Morin (1986). Sandhi in Walloon. In H. Andersen (ed.) Sandhi

Phenomena in the Languages of Europe. Mouton de Gruyter, Berlin. 453-474.

Ganong, W. (1980). Phonetic categorization in auditory word perception. Journal of

Experimental Psychology: Human Perception and Performance 6. 110-125.

Grammont, M. (1933). Traité de phonetique. Librairie Delagrave, Paris.

Haggard, M. (1978). The devoicing of voiced fricatives. Journal of Phonetics 6. 95-102.

Hahn, R. (1991). Spoken Uyghur. University of Washington Press, Seattle.

Harris, J. (2009). Why final devoicing is weakening. In K. Nasukawa & P. Backley

(eds.), Strength Relations in Phonology. Mouton de Gruyter, Berlin. 9-46.

Hayes, B. (1999). Phonetically-driven phonology: The role of Optimality Theory and

inductive grounding. In M. Darnell, E. Moravscik, M. Noonan, F. Newmeyer, and K.

Wheatly (eds.) Functionalism and Formalism in Linguistics, Volume I: General

Papers. John Benjamins, Amsterdam. 243-285.

Heim, M. (1976). Contemporary Czech. Slavica Publishers, Columbus, Ohio.

Hock, H. H. (1991). Principles of Historical Linguistics. 2nd ed. Mouton de Gruyter,

Berlin.

Hodson, B. and E. P. Paden (1981). Phonological processes which characterize

unintelligible and intelligible speech in early childhood. Journal of Speech and

Hearing Disorders 46. 369-373.

Hogan, J. and A. Roszypal (1980). Evaluation of vowel duration as a cue for the voicing

distinction in the following word-final consonant. Journal of the Acoustical Society of

America 67. 1764-1771.

Holmes, J. (1996). Losing voice: Is final devoicing a feature of Maori English? World

Englishes 15. 193-205.

Hualde, J. (1991). Basque Phonology. Routledge, London.

Hualde, J. (1992). Catalan. Routledge, London.

Hyman, L. (1976). Phonologization. In A. Juillard (ed.) Linguistic Studies Offered to

Joseph Greenberg, v. 2. Anma Libri, Saratoga, Ca., 407-418.

Ingram, D. (1989). Phonological Disability in Children. Whurr, London.

Iwasaki, S. and B. P. Ingkaphirom (2005). Thai. Cambridge University Press, Cambridge.

Jakobi, A. (1990). A Fur Grammar. Helmut Buske, Hamburg.

Jassem, W., and L. Richter (1989). Neutralization of voicing in Polish obstruents.

Journal of Phonetics 17. 317-325.

Jespersen, O. (1926). Lehrbuch der Phonetik. B. G. Truebner, Leipzig.

Jessen, M. (1998). Phonetics and Phonology of Tense and Lax Obstruents in German.

John Benjamins, Amsterdam.

Jessen, M. and C. Ringen (2002). Laryngeal features in German. Phonology 19. 189-218.

Jungraithmayr, H. (1970). Die Ron-Sprachen. Verlag J. J. Augustin, Glückstadt.

Kingston, J. and R. Diehl (1994). Phonetic knowledge. Language 70. 419-454.

Klatt, D. and L. Klatt (1990). Analysis, synthesis, and perception of voice quality

variations among female and male talkers. Journal of the Acoustical Society of

America 87. 820-857.

Kohler, K. (1979). Dimensions in the perception of fortis and lenis plosives. Phonetica

36. 332-343.

Krämer, M. (2000). Voicing alternations and underlying representations: The case of

Breton. Lingua 110. 639-663.

Kuzla, C., T. Cho, and M. Ernestus (2007). Prosodic strengthening of German fricatives

in duration and assimilatory devoicing. Journal of Phonetics 35. 301-320.

Labov, W. (2001). Principles of Linguistic Change: Social Factors. Blackwell, Malden

(Mass.)

Lehiste, I. (1970). Suprasegmentals. MIT Press, Cambridge.

Lehtonen, J. (1970). Aspects of Quantity in Standard Finnish. Jyväskylä University Press,

Jyväskylä, Finland.

Lightner, T. (1972). Problems in the Theory of Phonology. Vol. 1: Russian Phonology

and Turkish Phonology. Linguistic Research, Inc., Edmonton.

Lindblom, B. (1968). Temporal organization of syllable production. Speech

Transmission Laboratory Quarterly Progess and Status Report 2-3 (Royal Institute of

Technology, Stockholm). 1-6.

Lindblom, B. (1983). Economy of speech gestures. In P. MacNeilage (ed.) The

Production of Speech. Springer, New York. 217-245.

Lindblom, B. and M. Studdert-Kennedy (1967). On the role of formant transitions in

vowel recognition. Journal of the Acoustical Society of America 42. 830-843.

Lisker, L. (1957). Closure duration and the intervocalic voiced-voiceless distinction in

English. Language 33. 42-49.

Lisker, L. (1986). 'Voicing' in English: A catalogue of acoustic features signaling /b/ vs.

/p/ in trochees. Language and Speech 29. 3-11.

Lisker, L., A. Abramson, F. Cooper, and M. Schvey. (1969). Transillumination of the

larynx in running speech. Journal of the Acoustical Society of America 45. 1544-

1546.

Locke, J. (1983). Phonological Acquisition and Change. Academic Press, New York.

Lombardi, L. (1995). Laryngeal neutralization and syllable well-formedness. Natural

Language and Linguistic Theory 13. 39-74.

Lombardi, L. (1999). Positional faithfulness and voicing assimilation in Optimality

Theory. Natural Language and Linguistic Theory 17. 267-302.

Luelsdorf, P. (1975). A Segmental Phonology of Black English. Mouton, The Hague.

Major, R. and M. Faudree (1996). Markedness universals and the acquisition of voicing

contrasts by Korean speakers of English. Studies in Second Language Acquisition 18.

69-90.

Mann, V. and B. Repp (1980). Influence of vocalic context on perception of the [S]-[s]

distinction. Perception and Psychophysics 28, 213-228.

Michelson, K. (1988). A Comparative Study of Lake-Iroquoian Accent. Kluwer,

Dordrecht.

Montreuil, J.-P. (1999). The Romansch syllable. In H. van der Hulst and N. Ritter (eds.)

The Syllable: Views and Facts. Mouton de Gruyter, Berlin. 527-550.

Moreton, E. (2008). Analytic bias and phonological typology. Phonology 25. 83-127.

Myers, S. (2005). Vowel duration and neutralization of vowel length contrasts in

Kinyarwanda. Journal of Phonetics 33. 427-446.

Myers, S. and B. Hansen (2007). The origin of vowel length neutralization in final

position: Evidence from Finnish speakers. Natural Language and Linguistic Theory

25.157-193.

Newton, B. (1972). The Generative Interpretation of Dialect: A Study of Modern Greek

Phonology. Cambridge University Press, Cambridge.

Nivens, R. (1992). A lexical phonology of West Tarangan. In D. Burquest and W. Laidis

(eds.) Phonological Studies in Four Languages of Maluku. Summer Institute of

Linguistics, Arlington, Texas. 127-227.

Ohala, J. (1981). The listener as a source of sound change. In C. Masek, R. Hendrick and

M. Miller (ed.) Papers from the Parasession on Language and Behavior. Chicago

Linguistics Society, Chicago. 178-203.

Ohala, J. (1983). The origin of sound patterns in vocal tract constraints. In P. MacNeilage

(ed.) The Production of Speech. Springer, New York. 189-216.

Ohala, J. (1993). The phonetics of sound change. In C. Jones (ed.) Historical Linguistics:

Problems and Perspectives. Longman, London, 237-278.

Oller, D. (1973). The effect of position in utterance on speech segment duration in

English. Journal of the Acoustical Society of America 54. 1235-1247.

Oller, D. and B. Smith (1977). Effect of final syllable position on vowel duration in

infant babbling. Journal of the Acoustical Society of America 62. 994-997.

Oller, D., L. Wieman, W. Doyle and C. Ross (1976). Infant babbling and speech. Journal

of Child Language 3. 1-11.

Padgett, J. (2002). Russian voicing assimilation, final devoicing, and the problem of [v].

Unpublished paper available at http://people.ucsc.edu/~padgett/papers.html.

Passy, P. (1891). Étude sur les changements phonetiques et leurs caractères généraux.

Librairie Firmin-Didot, Paris.

Pierrehumbert, J. (2001). Exemplar dynamics: Word frequency, lenition and contrast. In

J. Bybee and P. Hopper (eds.) Frequency and the Emergence of Linguistic Structure.

John Benjamins, Amsterdam. 137-157.

Pind, J. (1986). The perception of quantity in Icelandic. Phonetica 43. 116-139.

Pitkin, H. (1984). Wintu Grammar. University of California Press, Berkeley.

Poppe, N. (1960). Buriat Grammar. Mouton, The Hague.

Port, R. and M. O'Dell (1985). Neutralization of syllable-final devoicing in German.

Journal of Phonetics 13, 455-471.

Raphael, L. (1971). Preceding vowel duration as a cue to the perception of the voicing

characteristics of word-final consonants in American English. Journal of the

Acoustical Society of America 51. 1296-1303.

Rubach, J. (1984). Cyclic and Lexical Phonology: The Structure of Polish. Foris,

Dordrecht.

Rubach, J. (1993). The Lexical Phonology of Slovak. Oxford University Press, Oxford.

Ruhlen, M. (1987). A Guide to the Languages of the World. Stanford University Press,

Stanford.

Sapir, E. (1990). Takelma Texts and Grammar (Collected Works of Edward Sapir VIII).

Mouton de Gruyter, Berlin.

Scatton, E. (1984). A Reference Grammar of Modern Bulgarian. Slavica Publishers,

Columbus, Ohio.

Selkirk, E. (1978). On prosodic structure and its relation to syntactic structure. In T.

Fretheim (ed.) Nordic Prosody II. TAPIR, Trondheim. 111-140.

Selkirk, E. (1986). On derived domains in sentence phonology. Phonology Yearbook 3.

371-405.

Shadle, C. (1997). The aerodynamics of speech. In W. Hardcastle and J. Laver (eds.) The

Handbook of Phonetic Sciences. Blackwell, Oxford. 33-64.

Sievers, E. (1901). Grundzüge der Phonetik. 5th ed. Breitkopf und Härtel, Leipzig.

Singh, C. Y. (2000). Manipuri Grammar. Rajesh Publications, New Delhi.

Slifka, J. (2006). Some physiological correlates to regular and irregular phonation at the

end of an utterance. Journal of Voice 20. 171-186.

Slowiaczek, L. and D. Dinnsen (1985). On the neutralizing status of Polish word-final

devoicing. Journal of Phonetics 13. 325-341.

Slowiaczek, L. and H. Szymanska (1989). Perception of word-final devoicing in Polish.

Journal of Phonetics 17. 205-212.

Smith, B. (1979). A phonetic analysis of consonantal devoicing in children's speech.

Journal of Child Language 6. 19-28.

Smith, C. (1997). The devoicing of /z/ in American English: Effects of local and prosodic

context. Journal of Phonetics 25. 471-500.

Smith, C. (1999). Marking the boundary: Utterance-final prosody in French questions and

statements. In J. Ohala, Y. Hasegawa, M. Ohala, D. Granville and A. Bailey (eds.),

Proceedings of the XIVth International Congress of Phonetic Sciences. Linguistics

Department, University of California: Berkeley. 1181-1184.

Smith, C. (2003). Vowel devoicing in contemporary French. French Language Studies

13. 177-194.

Smith, N. (1973). The Acquisition of Phonology. Cambridge University Press,

Cambridge.

Stevens, K., S. Blumstein, L. Glicksman, M. Burton, and K. Kurowski (1992). Acoustic

and perceptual characteristics of voicing in fricatives and fricative clusters. Journal of

the Acoustical Society of America 91. 2979-3000.

Sweet, H. (1877). Handbook of Phonetics. Oxford University Press, Oxford.

Thompson, L. (1965). A Vietnamese Grammar. University of Washington Press, Seattle.

Thurgood, G. and R. LaPolla (eds.) (2003). The Sino-Tibetan Languages. Routledge,

London.

Tucker, A. (1994). A Grammar of Kenya Luo (Dholuo). Rüddiger Koppe Verlag, Köln.

Turk, A. (1999). Structural influences on boundary-related lengthening in English. In J.

Ohala, Y. Hasegawa, M. Ohala, D. Granville and A. Bailey (eds.), Proceedings of the

XIVth International Congress of Phonetic Sciences. Linguistics Department,

University of California, Berkeley. 237-240.

Umeda, N. (1975). Vowel duration in American English. Journal of the Acoustical

Society of America 58. 434-445.

Velten, H. (1943). The growth of phonemic and lexical patterns in infant speech.

Language 19. 281-292.

Warner, N., Jongman, A., Sereno, J. and Kemps, R. (2004). Incomplete neutralization

and other sub-phonemic durational differences in production and perception:

Evidence from Dutch. Journal of Phonetics 32. 251-276.

Warren, R. (1970). Perceptual restoration of missing speech sounds. Science 167, 392-

393.

Wedel, A. and H. Van Volkinburg (2009). Modeling simultaneous convergence and

divergence of linguistic features between differently-identifying groups in contact.

Unpublished paper downloaded from

http://dingo.sbs.arizona.edu/~wedel/publications/.

Westbury, J. and P. Keating (1986). On the naturalness of stop consonant voicing.

Journal of Linguistics 22. 145-166.

Wetzels, W. and J. Mascaró (2001). The typology of voicing and devoicing. Language

77. 207-244.

Whitney, W. D. (1879) A Sanskrit Grammar. Breitkopf and Härtel, Leipzig.

Wolf, C. (1978). Voicing cues in English final stops. Journal of Phonetics 6. 299-309.

Wolfram, W. (1969). A Sociolinguistic Description of Detroit Negro Speech. Center for

Applied Linguistics, Washington, D.C.

Wolfram, W., and D. Christian (1976). Appalachian Speech. Center for Applied

Linguistics, Arlington, Va.

Wright, J. (1899). A Primer of the Gothic Language. Oxford University Press, Oxford.

Yu, A. (2004). Explaining final obstruent voicing in Lezgian: Phonetics and history.

Language 80. 73-97.